alt Hacker NewsI'm working on yet another computer vision graphical annotation tool — but with a twist. I have this wild idea that labels should live inside the image itself, stored as dedicated text chunks within the PNG file, completely eliminating the need for sidecar files.
The tool will support four annotation modes: Box, Polygon, Mask, and Keypoints — each with its own dedicated panel. You can switch modes by clicking the color-coded buttons on the toolbar, complete with smooth transitions. Labeling is a tedious task, so a bit of satisfying UI action here and there can't hurt.
It will also export labels to all major formats — and can (re)generate any sidecar file structure when needed.